Use multiple Cuda Stream in a GPU compute with Matlab ?

Hello, In my code Matlab, i have a for loop with inside a arrayfun to compute on the GPU device. I want to buy a new GPU card with the hyper-Q and cuda stream capability. So I have to know if it's possible to apply the multiple Cuda Stream to my code instead of the for loop to accelerate it.

Answer 1

[Matt J]

I'm pretty confident you could do it using a dedicated MEX file, which then calls a .cu file.

I don't think the Parallel Computing Toolbox let's you do it, however. Not as of R2013, anyway.

[Benjamin]

Thanks for the answer.

Actualy, today i use this code on a nvidia Tesla C1060 but it's very long to compute.

It's why we want a new GPU card, and we want fully use its power of compute.

My code is like that :

initialisation

for i=1:250

arrayfun(@function1,arg1,arg2,arg3)
arrayfun(@function2,arg1,arg2,arg3)

end

with arg1,arg2,arg3 are vector with a size of (62500,1)

since the new GPU card is much more power than the old one, i want to send to the GPU several iteration of the for loop in parallel.

Answer 2

[Dimitris Floros]

Just replying to see if you have you found a solution from within MATLAB environment. I am searching for something similar and I don't want to use .MEX because I will have to implement the functions outside MATLAB.